Department of Applied Mathematics, University of Venice WORKING PAPER SERIES Working Paper n. 203/2010 October 2010

Transcription

1 Department of Applied Mathematics, University of Venice WORKING PAPER SERIES Antonella Campana and Paola Ferretti Initial premium, aggregate claims and distortion risk measures in X L reinsurance with reinstatements Working Paper n. 203/2010 October 2010 ISSN:

2 This Working Paper is published under the auspices of the Department of Applied Mathematics of the Ca Foscari University of Venice. Opinions expressed herein are those of the authors and not those of the Department. The Working Paper series is designed to divulge preliminary or incomplete work, circulated to favour discussion and comments. Citation of this paper should consider its provisional nature.

3 Initial premium, aggregate claims and distortion risk measures in XL reinsurance with reinstatements Antonella Campana Dept. SEGeS University of Molise Paola Ferretti Dept. of Applied Mathematics and Advanced School of Economics University of Venice (October 2010) Abstract. With reference to risk adjusted premium principle, in this paper we study excess of loss reinsurance with reinstatements in the case in which the aggregate claims are generated by a discrete distribution. In particular, we focus our study on conditions ensuring feasibility of the initial premium, for example with reference to the limit on the payment of each claim. Comonotonic exchangeability shows the way forward to a more general definition of the initial premium: some properties characterizing the proposed premium are presented. Keywords: Excess of loss reinsurance; reinstatements; distortion risk measures; initial premium; exchangeability. JEL Classification Numbers: G22. MathSci Classification Numbers: 62P05. Correspondence to: Paola Ferretti Dept. of Applied Mathematics, University of Venice San Giobbe, Cannaregio Venezia, Italy Phone: [++39] (041) Fax: [++39] (041) ferretti@unive.it We are grateful to some MAF2010 conference members for suggestions and valuable comments on an earlier version of the paper. This research was partially supported by MIUR.

4 1 Introduction The excess of loss reinsurance model we study in this paper is related to the model that has been originally proposed and analyzed by Sundt [9] and that it has been subsequently generalized (see [1] and [3]). Given the initial premium P, the limit on the payment of each claim m, the number of reinstatements K (such that the aggregate limit M satisfies M = (K + 1)m), the aggregate deductible D, the percentages of reinstatement c i, (i = 1,..., K), the total premium income results to be a random variable, say δ(p ), which is so defined δ(p ) = P ( m K 1 i=0 c i+1 L X (D + im, D + (i + 1)m) where L X (a, a + b) = min{(x a) +, b} and (X a) + = X a if X > a, otherwise (X a) + = 0 (a, b R + ). From the point of view of the reinsurer, the aggregate claims S paid by the reinsurer for this XL reinsurance treaty, namely satisfy the relation S = S = L X (D, D + (K + 1)m) (2) K L X (D + im, D + (i + 1)m). (3) i=0 This reinsurance cover is called an XL reinsurance for the layer m xs d with aggregate deductible D and K reinstatements and provides total cover for the amount S. The initial premium P is supposed to cover the original layer L X (D, D + m) = min{(x D) +, m}. (4) The condition that the reinstatement is paid pro rata means that the premium for the i-th reinstatement is a random variable given by c i P m L X(D + (i 1)m, D + im) (5) where 0 c i 1 is the i-th percentage of reinstatement. As it is well-known, excess of loss reinsurance with reinstatement has been essentially studied in Actuarial Literature in the framework of collective model of risk theory and this choice requires the knowledge of the claim size distribution in the classical evaluation of pure premiums. Conversely, as a rule, only few characteristics of aggregate claims can be explained and the interest for general properties characterizing the involved premiums is still flourishing. In order to contribute to this research, we set our analysis in the framework of risk adjusted premiums (a more general choice than that of Walhin and Paris [10]) and we concentrate our attention on properties exhibited by initial premiums, both with respect to equilibrium condition (section 2), both with respect to feasibility with reference to the limit on the payment of each claim (section 3). Then the attention is moved to the recent framework of generalized risk adjusted premiums (Wu and Zhou [12]) where the main role is played by comonotonic exchangeability assumption: we propose the definition of the related generalized initial premium and we state some regularity properties displayed by it (section 4). Finally section 5 ends the paper with some concluding remarks. 1 ) (1)

5 2 Excess of loss reinsurance with reinstatements: some preliminary results Recently (see [3]) we studied the initial risk adjusted premium P as solution of the equilibrium condition expressing the fact that the value of the total premium income δ(p ) equals the distorted expected value of the aggregate claims S, that is where W g1 (δ(p )) = W g2 (S) (6) W gj (X) = 0 g j (S X (x))dx (7) and the functions g j (j = 1, 2) are distortion function, i.e. non-decreasing functions g j : [0, 1] [0, 1] such that g j (0) = 0 and g j (1) = 1; S X (x) = P (X > x) is the decumulative distribution function of X. This initial premium P is well-defined and it is given by P = K i=0 W g 2 (L X (im, (i + 1)m)) K 1 m i=0 c i+1 W g1 (L X (im, (i + 1)m)). (8) We see that (6) gives a sort of global equilibrium: the distortion risk measure associated with the aggregate claims to the layer S must be equal to the distortion risk measure associated with the total premium income δ(p ). If we apply the same scheme to each layer, we derive that the premium P i, for all i {0, 1,..., K}, must satisfy the following equation W g1 (P i ) = W g2 (L X (im, (i + 1)m)). (9) Given that distortion risk measures obey positive homogeneity, the initial premium P 0 is easily defined by the previous equation; in fact The premium for the i-th reinstatement is given by W g1 (P 0 ) = P 0 = W g2 (L X (0, m)). (10) P i = c ip 0 m L X(D + (i 1)m, D + im), (11) then we have c i P 0 m W g 1 (L X ((i 1)m, im)) = W g2 (L X (im, (i + 1)m)). (12) Hitherto we have implicitly assumed the reinstatements percentages c i to be known. If the reinstatement percentages c i are not fixed in advance, we obtain c i = mw g 2 (L X (im, (i + 1)m)) P 0 W g1 (L X ((i 1)m, im)) (13) where the initial premium P 0 given by (10). We note that both the initial premium P 0 both the percentages c i do not depend on the number of reinstatements K. The values 2

6 given by (13) are useful even when the c i are to be determined in advance. They will give us a point of reference in order to obtain local equilibrium for each reinstatement. If the reinstatements percentages c i are determined by formula (13), one can easily verify that the following equality holds P = P 0 both in the case of same distortion functions g 1 = g 2 (see [1]) both in the case of not necessarily equal distortion functions g j (j = 1, 2). In fact, by (8) and (13) it is K i=0 P = W g 2 (L X (im, (i + 1)m)) K 1 P 0 i=0 W g2 (L X ((i + 1)m, (i + 2)m)), (14) that is, and, finally, P = K i=0 W g 2 (L X (im, (i + 1)m)) 1 + K 1 i=0 W g2 (L X ((i+1)m,(i+2)m)) W g2 (L X (0,m)) (15) P = W g2 (L X (0, m)) = P 0. (16) Therefore local equilibrium condition for each reinstatement ensures global equilibrium as defined in (6). 3 Initial premiums and limit on the payment of each claim Generally, reinstatement percentages are fixed in advance: this assumption moves our attention to the study of the initial premium defined by local equilibrium condition (9). From now on we make the assumption that the premium for the i-th reinstatement (5) is a twopoint random variable distributed as c i P B pi and B pi denotes a Bernoulli random variable such that P r[b pi = 1] = p i = 1 P r[b pi = 0]. Closely related to equation (12) is the definition of the initial premium P 0i (i = 1,..., K) for which the local equilibrium condition is satisfied in each layer, that is P 0i = mw g 2 (L X (im, (i + 1)m)) c i W g1 (L X ((i 1)m, im)) = mg 2(p i+1 ) c i g 1 (p i ). (17) Clearly acceptability of the treaty in the market requires that P 0i m for each i = 1,..., K. In order to determine conditions ensuring the validity of this condition which is not generally true (see Campana [2]), we direct our attention to the analysis of the relation between two consecutive values of the initial premium, that is to the comparison of P 0i to P 0i+1. Let us consider the following real-valued function H defined on [0, p i+1 ]: H(x) = c i+1 g 1 (p i+1 )g 2 (p i+1 ) c i g 1 (p i )g 2 (x). Note that H(0) = c i+1 g 1 (p i+1 )g 2 (p i+1 ) 0. Given that H(p i+1 ) = g 2 (p i+1 )[c i+1 g 1 (p i+1 ) c i g 1 (p i )] g 2 (p i+1 )c i+1 [g 1 (p i+1 ) g 1 (p i )] 3

7 if the percentages of reinstatement are decreasing, then increasing monotonicity of the distortion function g ensures both non-positivity of H(p i+1 ) both decreasing monotonicity of H. If the distortion function g 2 is continuous, intermediate value theorem ensures that there exists at least one point x in [0, p i+1 ] such that As a consequence, we have that H(x) = c i+1 g 1 (p i+1 )g 2 (p i+1 ) c i g 1 (p i )g 2 (x) = 0. H(x) { 0, 0 x x; 0, x x p i+1. This means that, given any choice of probabilities p i+1 and p i (p i+1 p i ), it is possible to state that there exists al least one probability x [0, p i+1 ] such that P 0i+1 = P 0i. Moreover, we can conclude that it is possible to anticipate the relation underlying two consecutive initial premiums P 0i and P 0i+1 for any choice of a probability p i+2 such as follows P 0i+1 { P0i, 0 p i+2 x; P 0i, x p i+2 p i+1. (18) In this way, it is possible to define a finite sequence of initial premiums P 0i (i = 1,..., K) such that P 0i is increasing (decreasing): it is sufficient to choose a finite sequence of probabilities p 0i (i = 2,..., K + 1) where at any step, x p i p i 1, (respectively, 0 p i x). In this way we obtain the following results. Proposition 1 Given an XL reinsurance with K reinstatements and distortion functions g 1 and g 2, where g 2 is continuous and the percentages of reinstatement are decreasing, then there exists a finite sequence of probabilities p i (i = 1,..., K+1) such that the finite sequence of initial risk adjusted premiums P 0i (i = 1,..., K) is monotone (increasing or decreasing). Proposition 2 Given an XL reinsurance with K reinstatements and distortion functions g 1 and g 2, where g 2 is continuous and the percentages of reinstatement are decreasing, there exists a finite sequence of probabilities p i (i = 1,..., K + 1), with g 2 (p K+1 ) g 1 (p K ) or g 2 (p 2 ) g 1 (p 1 ), such that P 0i m, for all i = 1,..., K. 4 Generalized initial premiums and exchangeability As it is well-known, insurance premium principles and risk measures can be characterized with reference to a set of axioms, that is to a set of desirable properties that they should have for actuarial practice. Among them, recently Goovaerts et al. ([5], [6]), Wu and Zhou (see [12]) propose a generalization of Yaari s risk measure by relaxing his proposal. Comonotonic additivity has been generalized to comonotonic exchangeability in [5], while the link between comonotonic additivity and independent additivity has been addressed in [6]. Additivity for a finite number of comonotonic risks is substituted by countable additivity and countable exchangeability in [12] in order to characterize generalized distortion premium 4

8 principles. In this context, we refer to a non-empty collection of risks L on (Ω, F, P ) such that min{x, a} min{x, b}, ax and X + a are all in L for any X L and for any a, b R. In studying different characterizations of insurance premium principles and risk measures, Goovaerts et al. [5] refer to some realistic situations owing to specific sets of axioms: they refer to the notion of exchangeability, that with reference to a principle H states A1. Exchangeability H(X c + Y c ) = H(X c + Y c ) provided that H(X) = H(X ) (19) where the random vectors X c and Y c admit the same marginal distributions of X and Y, respectively, with the comonotonic dependence structure. The following theorem states a representation result of the so called generalized distortion principle Theorem 1 A premium principle H : L R + admits for all X L the representation ( ) H(X) = h g(s X (x))dx (20) 0 where h is a continuous non-decreasing function on R +, if and only if H satisfies exchangeability, monotonicity in stochastic order and continuity. As comonotonic additivity is the most essential property of distortion integral (and Choquet integral), comonotonic exchangeability by generalizing comonotonic additivity suggests a new definition of distortion premium principle (and Choquet price principle), that is ( ) Wg h (X) = h g(s X (x))dx (21) 0 which is called generalized risk adjusted premium principle; the function g is a distortion function and h is a continuous non-decreasing function on R +. Now we turn our attention to the analysis of local equilibrium condition (9) and in order to consider the case of generalized risk adjusted premium principle we define and study the following generalized local equilibrium condition W h 1 g 1 (P i ) = W h 2 g 2 (L X (im, (i + 1)m)) (22) where the functions g j are distortion functions and h j are continuous non-decreasing functions on R + (j = 1, 2). Accordingly, the definition of generalized initial premium naturally follows: it is solution, write P 0i (i = 1,..., K), of the generalized local equilibrium condition (22). Clearly, positivity of P 0i is assumed. In the following we focus our attention on properties exhibited by the generalized initial premium P 0i, with reference to some assumptions made on the functions involved, namely the distortion functions g j and the composing functions h j, (j = 1, 2). The analysis of equation (22) when aggregate claims are generated by a discrete distribution may be conducted by introducing the following related function [ F h 1h 2 ci x ] g 1 g 2 (x, p i, p i+1 ) = h 1 m g 1(p i ) h 2 [g 2 (p i+1 )]. (23) 5

9 In this way, the generalized initial premium P 0i is such that F h 1h 2 g 1 g 2 (P 0i, p i, p i+1 ) = 0. Then if g j (j = 1, 2) are continuously differentiable distortion functions and h j are continuously differentiable and non-decreasing functions on R + (j = 1, 2), where g 1 (p i ) 0 and h [ ci x 1 m g 1(p i ) ] 0, P 0i results to be a continuously differentiable function of (p i, p i+1 ). Moreover [ ] [ ] P 0i F h 1 h 2 g (p i, p i+1 ) = 1 g 2 (x, p i, p i+1 ) / F h 1 h 2 g 1 g 2 (x, p i, p i+1 ) = x g 1 (p i) p i p i x g 1 (p i ) P 0i p i+1 (p i, p i+1 ) = [ F h 1 h 2 g 1 g 2 (x, p i, p i+1 ) p i+1 = m h 2 [g 2(p i+1 )] g c i h 1 [ c 2 (p i+1) ix m g 1(p i )] g 1 (p i ) ] / [ F h 1 h 2 where x = P 0i. Then we can state the following result. g 1 g 2 (x, p i, p i+1 ) x Proposition 3 Let g j be continuously differentiable distortion functions and let h j be continuously differentiable and non-decreasing functions on R + (j = 1, 2), where g 1 (p i ) 0 and h [ ci x 1 m g 1(p i ) ] 0. Then the generalized initial premium P 0i (i = 1,..., K) is a continuously differentiable function of (p i, p i+1 ) with ( P 0i (p i, p i+1 ) = x g 1 (p i) g 1 (p i ), m h 2 [g 2(p i+1 )] g c i h 1 [ c 2 (p ) i+1) ix m g 1(p i )] g 1 (p i ) where x = P 0i. Note that the generalized initial premium P 0i is monotone function of the probabilities: as a function of the probability p i it is decreasing, while as a function of the probability p i+1 it is increasing. ] 6

10 5 Concluding remarks In the considered XL model of reinsurance, when the reinstatements are paid the total premium income becomes a random variable which is correlated to aggregate claims. Incomplete information on aggregate claims distribution amplifies the interest for general properties characterizing the involved premiums. In this paper our main goal is to study properties exhibited by initial premium: our analysis refers to two settings, that of risk adjusted premiums and that of generalized risk adjusted premiums. We assume that the premium for each reinstatement is a two-point random variable, a particularly interesting hypothesis given that the reinsurance companies often assess treaties under the conjecture that there are only total losses. This happens, for example, when they use the rate on line method to price catastrophe reinsurance. As a matter of fact, conditions ensuring feasibility of the initial premium with respect to comparison with the limit on the payment of each claim and regularity properties displayed by the generalized initial premium are presented. In particular, the regularity results of the generalized initial premium hint that further research may be addressed to the analysis of optimal initial premiums. References [1] Campana, A.: Reinstatement premiums for an excess of loss reinsurance contract. Giornale Istituto Italiano degli Attuari, LXIX, Roma (2006) [2] Campana, A.: Risk adjusted premiums for excess of loss reinsurance with reinstatements. Forthcoming on Mathematical Methods in Economics and Finance. [3] Campana, A., Ferretti, P.: What do distortion risk measures tell us on excess of loss reinsurance with reinstatements? Mathematical and Statistical Methods for Actuarial Sciences and Finance, Corazza, M., Pizzi, C. (Eds.), XII, Springer-Verlag (2010) [4] Dhaene, J., Vanduffel, S., Tang, Q.H., Goovaerts, M.J., Kaas, R., Vyncke, D.: Risk measures and comonotonicity: a review. Stochastic Models 22 (2006) [5] Goovaerts, M.J., Kaas, R., Dhane, J., Tang, Q.: Some new classes of consistent risk measures. Insurance: Mathematics and Economics 34 (2004) [6] Goovaerts, M.J., Kaas, R., Laeven, R., Tang, Q.: A comonotonic image of independence for additive risk measures. Insurance: Mathematics and Economics 35 (2004) [7] Hürlimann, W.: Excess of loss reinsurance with reinstatements revisited. ASTIN Bulletin 35(1) (2005) [8] Mata, A.J.: Pricing excess of loss reinsurance with reinstatements. ASTIN Bulletin 30(2) (2000)

11 [9] Sundt, B.: On excess of loss reinsurance with reinstatements. Bulletin of the Swiss Association of Actuaries, 1 (1991) [10] Walhin, J.F., Paris, J.: Excess of loss reinsurance with reinstatements: premium calculation and ruin probability of the cedent. Proceedings of the third IME Congress, London (1999). [11] Wang, S.: Premium calculation by transforming the layer premium density. ASTIN Bulletin 26(1) (1996) [12] Wu, X., Zhou, X.: A new characterization of distortion premiums via countable additivity for comonotonic risks. Insurance: Mathematics and Economics 38 (2006)